infer_oselm.py

'''

Create an ELM manually &
Train and test it on N-CMAPSS

'''
## Import libraries in python
import gc
import argparse
import os
import json
import logging
import sys
import h5py
import time
import matplotlib
import numpy as np
import pandas as pd
import seaborn as sns
from pandas import DataFrame
import matplotlib.pyplot as plt
from matplotlib import gridspec
import math
import random
from random import shuffle
from tqdm.keras import TqdmCallback

import importlib
from scipy.stats import randint, expon, uniform
import sklearn as sk
from sklearn import svm
from sklearn.utils import shuffle
from sklearn import metrics
from sklearn import preprocessing
from sklearn import pipeline
from sklearn.metrics import mean_squared_error
from math import sqrt
from tqdm import tqdm
import scipy.stats as stats
# from sklearn.utils.testing import ignore_warnings
# from sklearn.exceptions import ConvergenceWarning
# import keras

from utils.data_preparation_unit import df_all_creator, df_train_creator, df_test_creator, Input_Gen
from utils.dnn import one_dcnn, mlps
from utils.hpelm import ELM, HPELM

# import tensorflow.compat.v1 as tf
# tf.disable_v2_behavior()
seed = 0
random.seed(0)
np.random.seed(seed)
# Ignore tf err log
pd.options.mode.chained_assignment = None  # default='warn'


# from tensorflow.compat.v1 import ConfigProto
# from tensorflow.compat.v1 import InteractiveSession
# config = ConfigProto()
# config.gpu_options.allow_growth = True
# session = InteractiveSession(config=config)

#gpus = tf.config.experimental.list_physical_devices('GPU')
#for gpu in gpus:
#    tf.config.experimental.set_memory_growth(gpu, True)

# tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)
# tf.get_logger().setLevel(logging.ERROR)



# tf.config.set_visible_devices([], 'GPU')

current_dir = os.path.dirname(os.path.abspath(__file__))
data_filedir = os.path.join(current_dir, 'N-CMAPSS')
data_filepath = os.path.join(current_dir, 'N-CMAPSS', 'N-CMAPSS_DS02-006.h5')
sample_dir_path = os.path.join(data_filedir, 'Samples_whole')

model_temp_path = os.path.join(current_dir, 'Models', 'oned_fnn_rep.h5')
tf_temp_path = os.path.join(current_dir, 'TF_Model_tf')

pic_dir = os.path.join(current_dir, 'Figures')

'''
load array from npz files
'''
def load_part_array (sample_dir_path, unit_num, win_len, stride, part_num):
    filename =  'Unit%s_win%s_str%s_part%s.npz' %(str(int(unit_num)), win_len, stride, part_num)
    filepath =  os.path.join(sample_dir_path, filename)
    loaded = np.load(filepath)
    return loaded['sample'], loaded['label']

def load_part_array_merge (sample_dir_path, unit_num, win_len, win_stride, partition):
    sample_array_lst = []
    label_array_lst = []
    print ("Unit: ", unit_num)
    for part in range(partition):
      print ("Part.", part+1)
      sample_array, label_array = load_part_array (sample_dir_path, unit_num, win_len, win_stride, part+1)
      sample_array_lst.append(sample_array)
      label_array_lst.append(label_array)
    sample_array = np.dstack(sample_array_lst)
    label_array = np.concatenate(label_array_lst)
    sample_array = sample_array.transpose(2, 0, 1)
    print ("sample_array.shape", sample_array.shape)
    print ("label_array.shape", label_array.shape)
    return sample_array, label_array


def load_array (sample_dir_path, unit_num, win_len, stride):
    filename =  'Unit%s_win%s_str%s.npz' %(str(int(unit_num)), win_len, stride)
    filepath =  os.path.join(sample_dir_path, filename)
    loaded = np.load(filepath)

    return loaded['sample'].transpose(2, 0, 1), loaded['label']


def shuffle_array(sample_array, label_array):
    ind_list = list(range(len(sample_array)))
    print("ind_list befor: ", ind_list[:10])
    print("ind_list befor: ", ind_list[-10:])
    ind_list = shuffle(ind_list)
    print("ind_list after: ", ind_list[:10])
    print("ind_list after: ", ind_list[-10:])
    print("Shuffeling in progress")
    shuffle_sample = sample_array[ind_list, :, :]
    shuffle_label = label_array[ind_list,]
    return shuffle_sample, shuffle_label

def figsave(history, h1,h2,h3,h4, bs, lr, sub):
    fig_acc = plt.figure(figsize=(15, 8))
    plt.plot(history.history['loss'])
    plt.plot(history.history['val_loss'])
    plt.title('Training', fontsize=24)
    plt.ylabel('loss', fontdict={'fontsize': 18})
    plt.xlabel('epoch', fontdict={'fontsize': 18})
    plt.legend(['Training loss', 'Validation loss'], loc='upper left', fontsize=18)
    plt.show()
    print ("saving file:training loss figure")
    fig_acc.savefig(pic_dir + "/mlp_training_h1%s_h2%s_h3%s_h4%s_bs%s_sub%s_lr%s.png" %(int(h1), int(h2), int(h3), int(h4), int(bs), int(sub), str(lr)))
    return



def release_list(a):
   del a[:]
   del a

units_index_train = [2.0, 5.0, 10.0, 16.0, 18.0, 20.0]
units_index_test = [11.0, 14.0, 15.0]



def main():
    # current_dir = os.path.dirname(os.path.abspath(__file__))
    parser = argparse.ArgumentParser(description='sample creator')
    parser.add_argument('-w', type=int, default=50, help='sequence length', required=True)
    parser.add_argument('-s', type=int, default=1, help='stride of filter')
    parser.add_argument('-h1', type=int, default=2000, help='hidden1')
    parser.add_argument('-h2', type=int, default=2000, help='hidden2')
    parser.add_argument('-h3', type=int, default=1000, help='hidden3')
    parser.add_argument('-h4', type=int, default=500, help='hidden4')
    parser.add_argument('-bs', type=int, default=256, help='batch size')
    parser.add_argument('-ep', type=int, default=30, help='max epoch')
    parser.add_argument('-pt', type=int, default=20, help='patience')
    parser.add_argument('-vs', type=float, default=0.1, help='validation split')
    parser.add_argument('-lr', type=float, default=0.001, help='learning rate')
    parser.add_argument('-sub', type=int, default=1, help='subsampling stride')


    args = parser.parse_args()

    win_len = args.w
    win_stride = args.s
    partition = 3

    hidden1 = args.h1
    hidden2 = args.h2
    hidden3 = args.h3
    hidden4 = args.h4

    lr = args.lr
    bs = args.bs
    ep = args.ep
    pt = args.pt
    vs = args.vs
    sub = args.sub




    train_units_samples_lst =[]
    train_units_labels_lst = []

    for index in units_index_train:
        print("Load data index: ", index)
        sample_array, label_array = load_array (sample_dir_path, index, win_len, win_stride)
        #sample_array, label_array = shuffle_array(sample_array, label_array)
        print("sample_array.shape", sample_array.shape)
        print("label_array.shape", label_array.shape)
        sample_array = sample_array[::sub]
        label_array = label_array[::sub]

        sample_array = sample_array.astype(np.float32)
        label_array = label_array.astype(np.float32)

        print("sub sample_array.shape", sample_array.shape)
        print("sub label_array.shape", label_array.shape)
        train_units_samples_lst.append(sample_array)
        train_units_labels_lst.append(label_array)

    sample_array = np.concatenate(train_units_samples_lst)
    label_array = np.concatenate(train_units_labels_lst)
    print ("samples are aggregated")

    release_list(train_units_samples_lst)
    release_list(train_units_labels_lst)
    train_units_samples_lst =[]
    train_units_labels_lst = []
    print("Memory released")

    #sample_array, label_array = shuffle_array(sample_array, label_array)
    print("samples are shuffled")
    print("sample_array.shape", sample_array.shape)
    print("label_array.shape", label_array.shape)

    sample_array = sample_array.reshape(sample_array.shape[0], sample_array.shape[2])
    print("sample_array_reshape.shape", sample_array.shape)
    print("label_array_reshape.shape", label_array.shape)
    feat_len = sample_array.shape[1]
    print ("feat_len", feat_len)

    elm = HPELM(sample_array.shape[1], 1, accelerator="GPU", batch= 5000, norm=0.001  )
    elm.add_neurons(hidden1, "tanh")
    elm.add_neurons(hidden2, "tanh")
    elm.add_neurons(hidden3, "tanh")
    elm.add_neurons(hidden4, "tanh")

    elm.train(sample_array, label_array, "r")

    # Y = elm.predict(X)




    output_lst = []
    truth_lst = []

    for index in units_index_test:
        print ("test idx: ", index)
        sample_array, label_array = load_array(sample_dir_path, index, win_len, win_stride)
        # estimator = load_model(tf_temp_path, custom_objects={'rmse':rmse})
        print("sample_array.shape", sample_array.shape)
        print("label_array.shape", label_array.shape)
        sample_array = sample_array[::sub]
        label_array = label_array[::sub]
        print("sub sample_array.shape", sample_array.shape)
        print("sub label_array.shape", label_array.shape)
        sample_array = sample_array.reshape(sample_array.shape[0], sample_array.shape[2])
        print("sample_array_reshape.shape", sample_array.shape)
        print("label_array_reshape.shape", label_array.shape)

        sample_array = sample_array.astype(np.float32)
        label_array = label_array.astype(np.float32)

        # estimator = load_model(model_temp_path)

        y_pred_test = elm.predict(sample_array)
        output_lst.append(y_pred_test)
        truth_lst.append(label_array)

    print(output_lst[0].shape)
    print(truth_lst[0].shape)

    print(np.concatenate(output_lst).shape)
    print(np.concatenate(truth_lst).shape)

    output_array = np.concatenate(output_lst)[:, 0]
    trytg_array = np.concatenate(truth_lst)
    print(output_array.shape)
    print(trytg_array.shape)
    rms = sqrt(mean_squared_error(output_array, trytg_array))
    print(rms)
    rms = round(rms, 2)

    end = time.time()
    num_test = output_array.shape[0]

    for idx in range(len(units_index_test)):
        print(output_lst[idx])
        print(truth_lst[idx])
        fig_verify = plt.figure(figsize=(24, 10))
        plt.plot(output_lst[idx], color="green")
        plt.plot(truth_lst[idx], color="red", linewidth=2.0)
        plt.title('Unit%s inference' %str(int(units_index_test[idx])), fontsize=30)
        plt.yticks(fontsize=20)
        plt.xticks(fontsize=20)
        plt.ylabel('RUL', fontdict={'fontsize': 24})
        plt.xlabel('Timestamps', fontdict={'fontsize': 24})
        plt.legend(['Predicted', 'Truth'], loc='upper right', fontsize=28)
        plt.show()
        fig_verify.savefig(pic_dir + "/elm_unit%s_test_h1%s_h2%s_rmse-%s.png" %(str(int(units_index_test[idx])),
                                                                              int(hidden1), int(hidden2), str(rms)))



    print("Result in RMSE: ", rms)




if __name__ == '__main__':
    main()